In many embodiments of cluster systems, not all nodes of the cluster may be connected to external networks. For example, in asymmetric topologies or due to switch limitations including switch port unavailability or link aggregation (LAG) limitations, fewer than all of the cluster nodes may be connected to the upstream switch or router. Nodes serving as gateways to the cluster can manually route, encapsulate, and transmit packets to internal nodes, but this requires additional overhead processing and decreases efficiency.
In some embodiments, packets can be steered or transmitted for processing to an internal node based on transport layer information of the packet, such as source and/or destination port numbers. However, in embodiments in which packets arrive fragmented at the network layer, this transport layer information may not be available. Accordingly, packet fragments must be accumulated and assembled prior to steering. However, in many embodiments, a plurality of nodes may serve as gateways to the cluster, and accordingly, multiple nodes may receive fragments of a single packet. This may require significant amounts of inter-node traffic to coordinate the gateways, or may require designation of a single node for receiving all packets or packet fragments, creating a bottleneck and reducing efficiency of the cluster.